Root cause analysis of abnormalities in turbomachinery

Abstract

The method of the present invention enables determining the root cause of an anomaly in a turbomachine, and includes the steps of: a) receiving (220) measurement data relating to a time frame from a set of feature sensors installed on the turbomachine; b) receiving (230) an identifier of a target feature sensor whose measurement data shows an anomaly; c) receiving (240) a start time and an end time of a non-anomalous time subframe; d) receiving (250) a start time and an end time of an anomaly time subframe; e) receiving (260) identifiers of a plurality of feature sensors associated with features of the turbomachine that may be the root cause of the anomaly; and f) deriving (270) at least one feature of the turbomachine that may be the root cause of the anomaly based on a "contrastive analysis" of the "feature importance" values ​​of the feature sensors during the non-anomalous time subframes and the anomaly time subframes.

Description

[Technical Field]

[0001] The subject matter disclosed herein relates to root cause analysis of anomalies in turbomachinery. [Background technology]

[0002] Although turbomachinery is designed to avoid anomalies and maintenance is aimed at preventing them, anomalies occur in turbomachinery during its operation. An anomaly is a condition of the machine that deviates from standard operability; a first example of an anomaly is vibration at a particular location on the machine having a higher amplitude than the normal vibration amplitude at that location; a second example of an anomaly is rotational speed of a particular component of the machine having a higher than normal rotational speed amplitude for that component; a third example of an anomaly is temperature at a particular location on the machine having a higher value than the normal temperature at that location; a fourth example of an anomaly is pressure in a particular duct or cavity of the machine having a higher value than the normal pressure in that duct or cavity; and a fifth example of an anomaly is flow in a particular duct of the machine having a higher value than the normal flow in that duct or cavity. The expression "far from" should be interpreted to mean that the difference between a standard value, e.g., a rated value, and the actual value is greater than a predetermined difference, e.g., a predetermined percentage difference, which typically varies from parameter to parameter and may also vary, e.g., depending on the operating state of the machine.

[0003] Root cause analysis of anomalies in turbomachinery, i.e., finding out why an anomaly occurred in the past or is currently occurring in the turbomachinery, is very important to both manufacturers and users, but is very difficult to perform reliably. The complexity of turbomachinery, e.g., compressors or turbines, and the complexity of the applications in which the turbomachinery is installed and operates, e.g., oil and gas, make this task even more difficult. In some cases, it is difficult to even accurately identify the anomaly.

[0004] The technical literature discloses computer-implemented methods and computer-based systems aimed at fully automatic identification of machine anomalies. Some anomalies are relatively easy to identify; others are difficult to identify. To perform this effectively and reliably, a common feasible approach is to perform extensive testing and training on each machine in question. Typically, in-depth knowledge of the machine in question and the environment in which it is installed and operated is a major advantage for a reliable solution.

[0005] Similarly, the technical literature discloses computer-implemented methods and computer-based systems that aim to find the root cause of anomalies in machinery in a fully automated manner. Such a task is much more difficult and much more challenging to validate and reliable. Even in this case, extensive testing and training can be used to resolve the problem.

[0006] It is therefore desirable to have an easier approach to root cause analysis of anomalies in turbomachinery, particularly turbomachinery for oil and gas applications, without sacrificing effectiveness and reliability. Summary of the Invention

[0007] According to a first aspect, the subject matter disclosed herein relates to a computer-implemented method for root cause analysis of an anomaly in a turbomachine, the method including: a) receiving measurement data associated with a time frame from a set of feature sensors installed on the turbomachine; b) receiving an identifier of a target feature sensor in which an anomaly appears in the measurement data; c) receiving a start time and an end time of a non-anomaly time subframe in which no anomaly occurs; d) receiving a start time and an end time of an anomaly time subframe in which an anomaly occurs; e) receiving identifiers of a plurality of feature sensors associated with features of the turbomachine that may be a root cause of the anomaly; and f) deriving at least one feature of the turbomachine that may be a root cause of the anomaly based on a “contrastive analysis” of “feature importance” values ​​of the feature sensors in the non-anomaly time subframes and the anomaly time subframes. The terms “contrastive analysis” and “feature importance” are explained in the detailed description below.

[0008] According to other aspects, the subject matter disclosed herein relates to computer-based systems and turbomachine arrangements that implement such methods. [Brief explanation of the drawings]

[0009] A more complete understanding of the disclosed embodiments of this invention and many of the attendant advantages thereof will be readily obtained as the same become better understood by reference to the following detailed description when considered in connection with the accompanying drawings. [Figure 1] 1 illustrates a schematic block diagram of an embodiment of an innovative turbomachine arrangement including an innovative system. [Figure 2] 1 illustrates a flowchart of an embodiment of an innovative method for root cause analysis of anomalies in turbomachinery. [Figure 3] 3 shows a flowchart of a possible embodiment of certain steps of the method of FIG. 2; DETAILED DESCRIPTION OF THE INVENTION

[0010] As mentioned above, identifying anomalies in turbomachinery is difficult, and determining their root causes is even more difficult if reliable results are desired. Therefore, it is considered to limit the task to only a simplified, yet still difficult, problem, avoiding preliminary testing and training on one or more machines. This assumes that A) an anomaly is identified, i.e., the anomalous and non-anomalous periods are known, and B) the possible causes of the anomaly are known. The task is to select the cause with the greatest probability, i.e., the cause that is most likely to be the true root cause of the anomaly. Typically, the number of possible causes of the anomaly is large, e.g., from 10 to 100, and varies from anomaly to anomaly. For example, higher-than-normal vibrations may be caused by an abnormal flow value in one of a set of ducts, by an abnormal pressure value in one of a set of ducts, or by an abnormal rotational speed in one of a set of parts. The task is to select which of the duct flow rates, duct pressures, or part rotational speeds caused or is causing the identified anomaly at a particular time. In accordance with the subject matter disclosed herein, the problem is solved by a "control analysis," i.e., comparing abnormal and non-abnormal periods, without requiring prior knowledge of any abnormality, and especially without requiring prior training.

[0011] Next, embodiments of the present disclosure will be described in detail, examples of which are illustrated in the drawings. The examples and drawings are provided as an explanation of the present disclosure and should not be construed as limiting the present disclosure. Indeed, it will be apparent to those skilled in the art that various modifications and variations can be made to the present disclosure without departing from the scope or spirit of the present disclosure. In the following description, like reference numerals are used in the illustrative drawings of the embodiments to indicate elements that perform the same or similar functions. Moreover, for clarity of illustration, some reference numerals may not be repeated in all figures.

[0012] 1 shows, in a highly schematic manner, an embodiment of an innovative turbomachine arrangement 100, along with its user 10 interacting with arrangement 100. Arrangement 100 comprises a turbomachine 180 and an innovative computer-based system 140. User 10 may be an employee of the company that manufactured turbomachine 180, or an employee of a company responsible for testing turbomachine 180, or an employee of a company that manages the plant in which turbomachine 180 is installed. More generally, user 10 is a person or team involved in determining the root cause of an anomaly that has occurred or is occurring in turbomachine 180, a need that may be repeated from time to time with any new anomaly.

[0013] Arrangement 100 and its variants will be described in more detail below. It is important to note here that turbomachine 180 includes a set of characteristic sensors 182 installed on the machine, the number of sensors 182 being large, e.g., 100-1000, that repeatedly measure "characteristics", which may also be called "variables" of the turbomachine, e.g., temperature, pressure, volumetric and mass flow rates, displacement, speed (e.g., rotational speed), acceleration, vibration, valve opening levels, IGV set angular position, IGV detected angular position, gas composition, burner status, etc. Typically, sensors 182 are "real" sensors, i.e., devices that perform measurements inside the turbomachine and determine / output a signal (analog or digital), the amplitude of which corresponds to the measured value. Alternatively, in accordance with the subject matter disclosed herein, one or more of the sensors may be so-called "virtual" sensors; as is known, a "virtual" sensor is a piece of software running on a computer (which may be the same computer that runs the methods of the present invention) that, for example, repeatedly calculates a formula using data from one or more "real" sensors as input data to produce output data for the "virtual" sensors, as if the machine were equipped with the "real" sensors instead of the "virtual" sensors.

[0014] System 140 receives measurement data in some manner from sensor 182. Figure 1 shows an arrow between turbomachine 180 and system 140, which may be interpreted as a connection (wired or wireless) such that the measurement data is received directly from turbomachine 180. However, according to some embodiments, the measurement data may be collected by a computer system (not shown in Figure 1) at some time and then transferred to computer system 140 at a later time, for analysis only, through a computer connection (wired or wireless) or through a data storage device.

[0015] 2 shows a flowchart 200 of one embodiment of an innovative method for root cause analysis of anomalies in a turbomachine, such as turbomachine 180 of FIG. 1, which is a computer-implemented method that may be performed, for example, by computer system 140 of FIG. 1. A path in the flowchart is followed from start block 210 to stop block 280 for each identified anomaly, and it is generally expected that several anomalies will typically occur one after the other during operation of the turbomachine, and therefore may be repeated, for example, during investigation or inspection of the turbomachine (i.e., offline), or during testing of the turbomachine, or during operation of the turbomachine.

[0016] The innovative method is a) receiving data from a set of characteristic sensors installed on the turbomachine, the data corresponding to measurements performed within a time frame by the characteristic sensors of the set of characteristic sensors (block 220); b) receiving an identifier of a feature sensor from the set of feature sensors, where the anomaly appears in measurement data from at least this feature sensor, and where this feature sensor is a target feature sensor of the anomaly (block 230); c) receiving a first start time and a first end time of a first time subframe, the first time subframe being included within the time frame and no anomaly occurring during the first time subframe (block 240); d) receiving a second start time and a second end time of a second time subframe, the second time subframe being included within the time frame, and the anomaly occurring during the second time subframe (block 250); e) receiving identifiers of a plurality of characteristic sensors from a set of characteristic sensors associated with a characteristic of the turbomachine that may be a root cause of the anomaly (block 260); f) deriving at least one feature of the turbomachinery that is considered to be a root cause of the anomaly based on a “contrastive analysis” of “feature importance” values ​​of feature sensors among the plurality of feature sensors during the first time subframe and the second time subframe (block 270).

[0017] The "contrastive analysis" in step "f" refers to comparing anomalous and non-anomalous periods, particularly comparing the "feature importance" of anomalous and non-anomalous periods. Step "f" is described further below with reference to Figure 3. The term "feature importance" refers to the degree or level of influence of an input feature (i.e., variable) on an output feature (i.e., variable). If a system is considered a black box with several input variables and one output variable, any particular output value can be considered based on the influence of all input values, although each input value may have a different contribution to the particular output value. A system can be associated with a "prediction model," which is usually very complex, and an "explanation model," which should be very simple for easy understanding. A very effective type of "explanation model" is a linear function of binary variables that can implement an "additive feature attribution method." Further details on this topic can be found, for example, in the paper "A unified approach to interpreting model predictions" by Scott M. Lundberg and Su-In Lee, Proceedings of the 31st International Conference on Neural Information Processing Systems - NIPS 2017.

[0018] Specifically, in step "f", at least one feature is selected from the plurality of features, i.e., the features associated with the plurality of feature sensors mentioned in step "e". In particular, in step "f", a feature importance difference value of each feature sensor of the plurality of feature sensors is calculated, and a highest feature importance difference value is determined therefrom. The reason why more than one feature may be obtained in step "f", as will be explained later, is related to the fact that the innovative method can be used as an aid to a human (or a team of humans, e.g., technical experts), so that the final determination of the true root cause may, for example, be left to such a human (or a team of humans, e.g., technical experts) also based on their expertise, or may require further testing and / or investigation.

[0019] The "control analysis" of step "f" according to the presently disclosed subject matter comprises: - calculating, during the first time subframe and during the second time subframe, a feature importance value for each feature sensor of the plurality of feature sensors, in particular as a contribution of the feature to the regression of the target variable; calculating a feature importance difference value for each feature sensor of the plurality of feature sensors, the feature importance difference value being a difference between a feature importance value in the first temporal sub-frame and a feature importance value in the second temporal sub-frame; - determining the highest or greatest feature importance difference value among the calculated feature importance difference values.

[0020] Advantageously, the feature importance values ​​are calculated based on a model (typically an explanatory model) of the turbomachine that is a linear function of binary variables, the binary variables corresponding to the turbomachine features corresponding to the plurality of feature sensors, as described, for example, in the article by Scott M. Lundberg and Su-In Lee, cited above.

[0021] According to some exemplary embodiments, the "receiving" step, i.e., one or more or all of steps "b," "c," "d," and "e," includes receiving input from a user, e.g., user 10 of Figure 1. This is particularly true for steps "b" and "e," i.e., identifying target features and identifying possible root cause features.

[0022] According to some exemplary embodiments, the occurrence of an anomaly is determined by a human based on human observation of the turbomachine, for example, measurement data thereof.

[0023] Typically, the second time subframe, or "anomalous" time subframe, follows the first time subframe, or "non-anomalous" time subframe. In reality, the anomaly may have begun at the end of the first time subframe, but its effect may not be evident during or at the end of the first time subframe. In the case of a human error, even if only a small amount of measurement data is collected within the first time subframe while the anomaly is occurring, the innovative method still provides good results.

[0024] In steps "b" and "e" above, reference is made to "identifiers" as a means for identifying the characteristics of the sensors and turbomachinery.

[0025] In steps "c" and "d" above, reference is made to a "start time" and an "end time" as a means of identifying a "temporal sub-frame." Similarly, a "temporal sub-frame" may be identified by, for example, a "start time" and a "duration," or an "end time" and a "duration."

[0026] According to an advantageous embodiment, for example the embodiment of FIG. 3, step "f" comprises: f1) creating a model (typically an explanatory model) of the turbomachine based on the received measurement data (particularly the measurement data in the first time subframe, i.e., the "non-anomalous" time subframe), the model having as inputs turbomachine features corresponding to at least a plurality of feature sensors and as outputs turbomachine features corresponding to at least a target feature sensor (block 272); f2) calculating a feature importance value for each feature sensor of the plurality of feature sensors separately in the first time subframe and the second time subframe with respect to measurement data from at least the target feature sensor based on the model created in substep "f1" (block 274); f3) calculating a feature importance difference value for each feature sensor of the plurality of feature sensors based on the feature importance values ​​calculated in sub-step "f2", the feature importance difference value being the difference between the feature importance value in the first time sub-frame and the feature importance value in the second time sub-frame (block 275); f4) determining the highest feature importance difference value among the feature importance difference values ​​calculated in substep "f3" (block 276); f5) Deriving associated features of the turbomachine that are considered to be the root cause of the anomaly from the highest feature importance difference value determined in substep "f4" (block 278).

[0027] The model in sub-step "f1" is advantageously a regression model and more advantageously may be implemented by a recurrent neural network, in particular a neural network of the LSTM (= "long short-term memory") type. Such a model, in particular a neural network, is trained to predict target features based on measurement data of input feature sensors using the data received in step "a".

[0028] According to an advantageous embodiment, sub-step "f2" is implemented by an explainability method applied to the model created in sub-step "f1" and based on Shapley value-related techniques, in particular the SHAP value (see, for example, the previously cited paper by Scott M. Lundberg and Su-In Lee, which provides a general description of Shapley value-related techniques and a specific description of the SHAP value). Such explainability methods arise from game theory and essentially assign each feature a value corresponding to the expected change in model prediction when conditioned on that feature. The basic procedure behind such explainability methods is to retrain the model on all possible feature subsets S of F (where F is the set of all features) and assign to each feature an importance value that represents the impact of including that feature in the model prediction. To calculate this impact, a model is trained with the feature and another model is trained without the feature. Since the impact of not including a feature depends on other features in the model (collinearity effect), the aforementioned difference is calculated for all possible feature subsets. The Shapley value is then calculated and used as the feature attribution. These are weighted averages of all possible differences between predictions obtained with a feature and predictions obtained without a feature. Usually, this calculation is approximated to speed up the procedure.

[0029] Preferably, the operating conditions of the turbomachine in the first time subframe and the operating conditions of the turbomachine in the second time subframe are similar. Similarity may be based on the values ​​of input features; for example, similar conditions for a compressor may mean being within the same rotational speed range, and / or the same suction pressure range, and / or the same discharge pressure range. Similarity may also be based on the values, i.e., influence, of output features; similar conditions may mean that, in the absence of anomalies, the target feature has the same value or is within the same value range. From a practical perspective, similarity may be based on temporal proximity; similarity may be present if the first and second time subframes are consecutive or temporally close to each other (e.g., less than 10%, 20%, 50%, or 100% of the time distance of the first or second time subframe).

[0030] According to some advantageous embodiments, In substep "f4", a set of highest feature importance difference values ​​is determined; In sub-step "f5", the feature importance difference values ​​from the set of highest feature importance difference values ​​are ranked and the corresponding associated turbomachine features are ranked as root causes of the anomaly.

[0031] In this way, several features are offered as possible root causes of the anomaly, ordered according to their probability of being the true root cause.

[0032] According to some advantageous embodiments, in sub-step "f5", a confidence value is determined to determine the root cause of the anomaly. In particular, sub-steps "f1", "f2", "f3", "f4", and "f5" are repeated based on different initializations of the neural network weights, and for each iteration a separate feature ranking is obtained, and for each feature the mean value of its ranking position and the variance of its ranking position are determined, and the confidence of the feature that is the root cause of the analysis is the inverse of its variance.

[0033] The method for root cause analysis of anomalies in turbomachinery described and claimed herein can be implemented by a computer-based system, such as system 140 of FIG. 1 , configured to execute the method. The system may essentially comprise a processor, such as processor 142 of FIG. 1 , a memory, such as memory 146 of FIG. 1 , coupled to processor 142 and configured to store programs and data, and a human I / O interface, such as human I / O interface 144, coupled to processor 142. These components 142, 144, and 146 are key components of a computer; thus, system 140 may be, for example, a so-called “workstation” or a so-called “server,” or a so-called “cluster” of computers. To execute the innovative method, a suitable computer program is stored in the memory. To execute the method, input from user 10 is received from the human I / O interface and transmitted to the processor. As previously described, the innovative system is configured to receive measurement data from the turbomachinery in some manner (e.g., see the arrows in FIG. 1 ). A typical possibility is that the system 140 includes a database 148 for storing data, in particular measurement data, from one or more turbomachines. Transferring measurement data from a turbomachine to a computer located remotely from the turbomachine and storing the measurement data in a database located within or coupled to the computer is known in the art and is outside the scope of protection of the present patent application.

[0034] According to some embodiments, the innovative system is configured to perform the innovative method offline, in other words, the measurement data may have been transferred from the turbomachine to the database well in advance (e.g., an hour, a day, or a month) before being processed according to the methods disclosed herein.

[0035] According to another embodiment, an innovative system, such as system 140 of FIG. 1, is configured to perform an innovative method during operation of a turbomachine (or multiple turbomachines), such as turbomachine 180 of FIG. 1.

[0036] According to possible embodiments, an innovative system, e.g., system 140 of FIG. 1, may be configured to automatically identify one or more anomalies (not necessarily all) or one or more anomaly types (not necessarily all) in a turbomachine, e.g., turbomachine 180 of FIG. 1. For example, suitable software may be stored in a memory of the system, e.g., memory 146 of FIG. 1, to perform such tasks and provide information to software implementing the innovative method. The suitable software may also identify "anomalous" and "non-anomalous" time subframes and provide them to software implementing the innovative method.

[0037] As shown in Figure 1, the innovative system may be integrated into a turbomachine arrangement, such as arrangement 100 of Figure 1. Such a system essentially comprises a turbomachine, such as turbomachine 180 of Figure 1, and an innovative system, such as system 140 of Figure 1.

[0038] Understanding the root cause of an anomaly in a turbomachine is useful not only for gaining more / better knowledge of this turbomachine, but also for better designing this or similar turbomachines (especially one or more turbomachine components). Indeed, based on the identified root cause(s), according to some embodiments, it is possible to trigger an alert / alarm (audible or visible) to a user when the turbomachine is operating (in the field or during testing) and / or take some action, e.g., via a computer controlling on the turbomachine or on one or more subsystems of the turbomachine or one or more subsystems coupled to the turbomachine. Actions may be motivated, for example, by safety concerns or efficiency objectives. An action may be, for example, shutting down the turbomachine, or activating a security system or subsystem, or changing the adjustment of a component (e.g., opening or closing a valve).

Claims

1. 1. A computer-implemented method for root cause analysis of anomalies in turbomachinery, the method comprising: a) receiving (220) data from a set of characteristic sensors installed on the turbomachine, the data corresponding to measurements performed within a time frame by the characteristic sensors of the set of characteristic sensors; b) receiving (230) an identifier of a feature sensor from the set of feature sensors, the anomaly appearing in measurement data from at least the feature sensor, the feature sensor being a target feature sensor for the anomaly; c) receiving (240) a first start time and a first end time of a first time subframe, the first time subframe being included within the time frame and the anomaly not occurring during the first time subframe; d) receiving (250) a second start time and a second end time of a second time subframe, the second time subframe being included within the time frame and the anomaly occurring during the second time subframe; e) receiving (260) identifiers of a plurality of characteristic sensors of the set of characteristic sensors associated with a characteristic of the turbomachine that may be a root cause of the anomaly; f) deriving (270) at least one feature of the turbomachine that is likely a root cause of the anomaly based on a contrast analysis of feature importance values ​​of the feature sensors of the plurality of feature sensors during the first time subframe and the second time subframe; said control analysis in step "f" (270) - calculating a feature importance value for each feature sensor of said plurality of feature sensors during said first time sub-frame and during said second time sub-frame; - calculating a feature importance difference value for each feature sensor of the plurality of feature sensors, the feature importance difference value being the difference between the feature importance value in the first temporal sub-frame and the feature importance value in the second temporal sub-frame; - determining the highest or greatest feature importance difference value among said calculated feature importance difference values.

2. The method of claim 1 , wherein feature importance values ​​are calculated based on a model of the turbomachine that is a linear function of binary variables, the binary variables corresponding to features of the turbomachine that correspond to the plurality of feature sensors.

3. Step f is f1) creating (272) a model of the turbomachine based on the measurement data, the model having as inputs characteristics of the turbomachine corresponding to at least the plurality of characteristic sensors and as outputs characteristics of the turbomachine corresponding to at least the target characteristic sensor; f2) calculating (274) a feature importance value for each feature sensor of the plurality of feature sensors during the first time sub-frame and the second time sub-frame with respect to measurement data from at least the target feature sensor based on the model created in sub-step "f1"; f3) calculating (275) a feature importance difference value for each feature sensor of the plurality of feature sensors based on the feature importance values ​​calculated in sub-step "f2", the feature importance difference value being the difference between the feature importance value in the first temporal sub-frame and the feature importance value in the second temporal sub-frame; f4) determining (276) the highest feature importance difference value among the feature importance difference values ​​calculated in substep "f3"; and f5) deriving (278) associated features of the turbomachine that are considered to be root causes of the anomaly from the highest feature importance difference value determined in sub-step "f4".

4. 4. The method of claim 3, wherein the model in sub-step "f1" (272) is a regression model.

5. 5. The method of claim 4, wherein the model in sub-step "f1" (272) is implemented by a recurrent neural network.

6. 6. The method of claim 5, wherein the model in sub-step "f1" (272) is implemented by a neural network of the LSTM type.

7. 2. The method of claim 1, wherein sub-step "f2" (274) is performed by a Shapley value related technique.

8. The method of claim 1 , wherein operating conditions of the turbomachine in the first time subframe and operating conditions of the turbomachine in the second time subframe are similar.

9. In substep f4 (276), a set of highest feature importance difference values ​​is determined; 4. The method of claim 3, wherein in substep f5 (278), the feature importance difference values ​​among the set of highest feature importance difference values ​​are ranked and correspondingly associated features of the turbomachine are ranked as root causes of the anomaly.

10. 4. The method of claim 3, wherein in substep f5 (278) a confidence value is determined to determine the root cause of the anomaly.

11. A computer-based system (140) configured to perform the root cause analysis method of claim 1.

12. The computer-based system (140) of claim 11 configured to perform the method of claim 1 during operation of the turbomachine (180).

13. The computer-based system (140) of claim 12, configured to trigger an alert / alarm based on a root cause identified through said method.

14. 13. The computer-based system of claim 12, configured to take action on the turbomachine or on one or more subsystems of or coupled to the turbomachine based on root causes identified through the method.

15. A turbomachine arrangement (100) comprising a turbomachine (180) and a system (140) according to claim 11.

16. A turbomachine arrangement (100) comprising a turbomachine (180) and a system (140) according to claim 12.

Images